基于CSP和RR的多类运动想象脑电信号的识别分类研究

Recognition and Classification of Multi-class Motor Imagery EEG Signals Based on CSP and RR

脑-机接口通过大脑皮层的EEG活动或者大脑里单个神经元活动使得用户可以来控制设备。这方面最具挑战性的问题之一就是如何提高脑电信号的识别精度。本文采用少通道以及共同空间模式-岭回归分析的模式识别方法,并将其应用到四种运动想象脑电的识别分类。首先对原始数据进行有效的预处理,有漂移矫正,滤波,改进的ICA(Independent Component Analysis)去除伪迹;再利用CSP(Common Space Pattern)和HHT(Hibert-Huang Transform)分别对预处理好的数据进行特征提取;最后再将两种算法提取的特征分别进行SVM(Support vector machine),LDA(Linear Discriminant Analysis)和RR(Ridge Regression)进行分类。实验结果证明,共同空间模式-岭回归分析最后的分类效果是最好的,平均分类识别率约为82.93%,数据中9名被试的最高和最低的分类识别率之间的标准差为1.37%。

Users are equipped with the capability of controlling devices through EEG activity of cerebral cortex and one single neuronal activity in the brain, based on the brain-computer interface. In this regard, the most chal-lenging problem lies in how to improve the identification precision of electroencephalogram signals. The paper adopts an identification method called fewer channels and common spaces mode-ridge regression analysis, to iden-tify and categorize four motor imagination EGGs. Original data is pretreated effectively at first, and then drift cor-rection, filtration and modified ICA （Independent Component Analysis） are undertaken to remove artifacts; CSP （Common Space Pattern） and HHT （Hibert-Huang Transform） are used to extract features of pretreated data; The features extracted by two algorithms are categorized in terms of SVM （Support vector machine）, LDA （Linear Dis-criminant Analysis） and RR （Ridge Regression） in the end. Experimental results show that common spaces mode-ridge regression analysis is the most effective tool when it comes to categorization, with an average identifi-cation rate of almost 82.93%. The standard deviation between the highest identification rate and the lowest identifi-cation reaches 1.37% out of nine samples to be categorized.